Calibrating the model

We want to build a model of California DTS fishery. Most of the parameters of the model are fixed by data (all the biological side and most of the fleet side). What we need to calibrate is:

The way I see it there are three ways to accomplish this:

  1. Use logbook data:
    1. Fix behaviour to logbook data
    2. Change catchability/hold size to minimise distance from aggregates
    3. Now vary behavioural parameters and minimise distance from logbook fits
  2. Assume profit maximization:
    1. Start with some adaptive behaviour
    2. Change catchability/hold size to minimise distance from aggregates
    3. Change behavioural parameters to maximize profits
    4. Loop through 2 and 3 until convergence
  3. Assume simple behaviour
    1. Fix the adaptive behaviour (say EEI)
    2. Change catchability/hold size to minimise distance from aggregates

While 3 is obviously the worst of the options it is also the quickest and we can do it immediately to study the model and its policy suggestions. I do this now because I think this functions quite well as a possible “policy-relevant” deliverable and we can use this to benchmark the benefit of having logbook data.

The calibration is not too hard. We can’t use landings per se because landings are constrained by the ITQ system, so instead we try to calibrate the model so that the quota’s “attainment rates” (we get these from the Wilderness Markets report) and average profits (reported by the catcher vessel report) are more or less correct.

Calibration per se is just another Bayesian optimisation. It is instructive to keep catchability equal for all species to see how it relates to profits when hold sizes change. The Bayesian posterior we generate shows that there is a curve connecting holdsize and catchability so that we can achieve more or less our average profit target by staying on this curve

However the catch composition with catchability equal to all species is not great. We get sablefish and dover sole about right (which is important because they are the major catches) but we overfish thornyheads and underfish yelloweye (although that’s almost unavoidable).

So if we assume that boats’ hold sizes are more or less 8T, then we can start modifying catchability (making it heterogenous per species) around the value given by the curve (0.000290 for 8T boats).
We use the resulting catchabilities as our model.

Winners / Losers

One easy thing we can do is of course to study how profits are allocated between each port on average when the ITQ system is put in place. While the simulated profits average 150,000$ (about 40,000$ more than the empirical average) they are dispersed quite unevenly with fishers from some ports making consistent losses.

A nice validation comes from comparing these simulated profits with the exit rate from each port observed empirically by 2014. Ports where simulated profits are the lowest are also the ones that saw the highest percentage of exits. Because our model simulates no exit and no transfer we can’t account for people moving into Fort Bragg and if we exclude it the correlation between simulated profits and exits is of .83 (and .28 with Fort Bragg included) which is very heartening since we didn’t calibrate against it.

Since we haven’t modelled the difference in size of the boats our model suggests that the advantage of fishers in the Oregon area is fundamentally geographical rather than due to better capital.

Comparative Statics

We can study also how windfalls are redistributed per port. Let’s take again the example of a doubling of sablefish prices. The next figure shows that while everybody would benefit, the Oregon ports would be the ones gaining more.

Fixed Gear Catchers

One question that came up from ITQ groundfish review was the disparity between trawlers and the boats who switched to fixed gear targeting primarily sablefish.
Our model can simulate gear switching, so we can study this phenomenon in some detail. The main difficulty here is that I’d need catchability/selectivity/retention for fixed gear some of which would have to be calibrated on the side. For the sake of speed here, I am just going to assume that fixed gear catches sablefish at about the same rate as trawlers but catches no other species.

If there is a 5% probability for any fisher to switch gear at random (plus the usual imitation uptake) would we see people switching dramatically to fixed gear? The answer in the simulation is no. The effect is very minor. A few fishers always switch but it never takes hold.

Because the number of fixed gear boats simulated is so small, there is no real effect on the ITQ prices for sablefish.

Optimal Policies

One thing we want to do of course it simulate policy to provide suggestions. Always with an eye toward suggest simpler, faster rules. Now, for California we might ask ourselves if it is possible to substitute the complicated many-species ITQ with something simpler. Here we try 3 different things:

  1. Fixed Season Length for the next 65 years
  2. Fixed TAC on Dover Sole for the next 65 years
  3. Fixed ITQ on Dover Sole for the next 65 years

What is interesting about this is that ITQ performs much worse than TAC or Season Length. This reminds me very much of the second best literature in Economics: there is a best policy out there, but if any one condition for it fails then the second best policy (the one that is actually attainable) is not “something as similar as possible” to the first best policy but might be something very different. For example, 5 species ITQ system might be “best” but if you only have one species you can control, you might be better off not doing an ITQ at all, like in this case.

First let’s take a look at the average profits made for each policy, each year (all these plots are averages from 50 separate simulations):

The ITQ system performs very poorly (the quota numbers and the season length are set by separate Bayesian optimisations). The reason it performs so poorly is ironically the same reason it tends to perform well in the abstract case: ITQ provide incentives to people to avoid protected fish. So that boats will try their best to avoid catching Dover Sole, even if it means giving up on good fishing spots.
We can observe this by looking at their catches per unit of effort. Here I show the catches of Dover Sole per unit of effort and the catches of all other species per unit of effort. It is clear that the ITQ pushes people to avoid Dover Sole as much as possible especially the first few years (as the difference is much larger for dover sole per unit of effort than it is for all the other landings combined).

The need to avoid Dover Sole emerges from the ITQ system in the model. If we look at the average quota prices we see that Dover Sole quotas sell for higher than the sale price (0.66 $/kg represented by the orange line) when we impose our Dover-Sole only policy.